Power Platform prompts move to newer GPT models by 2 December

Power Platform prompts move to newer GPT models by 2 December

3 November 2026: Microsoft is retiring the models behind every prompt in Power Apps, Power Automate and Copilot Studio that still runs on an older GPT, Claude or Grok model, and will move those prompts to three newer models automatically by 2 December 2026. Apps, flows and agents that call the prompts move with them, and Microsoft warns that their answers may change.

Not yet independently verified. We read the notice in the Message Center archive at mc.merill.net, which mirrors Microsoft's text; we have not yet found a Microsoft-hosted page with the same model list. We will update this when it can be confirmed, and remove this note.

Which models go and what replaces them

The notice sorts the retirements by the three prompt tiers makers choose from:

• Basic: GPT-4.1 mini is replaced by GPT-6 Luna.
• Standard: GPT-4.1, GPT-5 chat, GPT-5.3 chat, Claude Sonnet 4.6 and Grok 4.1 Fast reasoning are replaced by ChatGPT-latest.
• Premium: GPT-5 reasoning, GPT-5.2 reasoning and Claude Opus 4.6 are replaced by GPT-6 Reasoning.

The standard tier carries the widest change. Five models from three vendors collapse into one, so a team that chose a Claude or Grok model for a particular extraction or classification prompt loses that choice, not just a version number. Microsoft names no earlier phase: the affected prompts are fully retired by 2 December, and the automatic migration completes by the same date.

The same price per tier, a different bill

Microsoft says tier pricing does not change. It also says the actual cost of a prompt can vary with the number of tokens the new model reads and writes, so a flow that runs a prompt thousands of times a day may consume credits at a different rate after the switch even though no price list moved. Reasoning models in particular tend to produce longer outputs.

No action is needed to trigger the migration. The risk is in what the migrated prompt returns: a prompt tuned to produce a fixed JSON shape, a short label or a yes/no answer for the next step of a flow can start returning something the next step does not expect.

Testing before the switch

Microsoft's recommended steps are concrete enough to plan around. Find the prompts on retiring models, run each against its replacement, adjust the instructions where the output changes, tell the makers who own them, and watch results and consumption after the move. Two points matter most in practice:

1. Flows that parse a prompt's output (a JSON field, a category name) should be tested with real inputs, because a format change breaks the flow rather than producing a visibly wrong answer.
2. Agents in Copilot Studio that use a premium prompt for multi-step reasoning are the ones most likely to change behaviour, since three reasoning models from two vendors become one.

Prompts on models not named in the notice are untouched by this change.

Our catalogue lists 20 robotic process automation products and 11 low-code development platforms, the categories most Power Platform estates are compared against.

Related on TrustList:

• Power Platform cuts off old Azure DevOps connections by 30 November

Sources

• Microsoft 365 Message Center archive (merill.net): MC1536101 Power Platform, newer GPT models replace Anthropic, Grok, and earlier GPT models, 3 November 2026